CN115050386B - Automatic detection and extraction method for whistle signal of Chinese white dolphin - Google Patents
Automatic detection and extraction method for whistle signal of Chinese white dolphin Download PDFInfo
- Publication number
- CN115050386B CN115050386B CN202210541058.1A CN202210541058A CN115050386B CN 115050386 B CN115050386 B CN 115050386B CN 202210541058 A CN202210541058 A CN 202210541058A CN 115050386 B CN115050386 B CN 115050386B
- Authority
- CN
- China
- Prior art keywords
- time
- signal
- cross
- chinese white
- whistle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 241001479123 Sousa chinensis Species 0.000 title claims abstract description 64
- 238000001514 detection method Methods 0.000 title claims abstract description 18
- 238000000605 extraction Methods 0.000 title claims abstract description 14
- 238000010586 diagram Methods 0.000 claims abstract description 30
- 241001481833 Coryphaena hippurus Species 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 60
- 239000011159 matrix material Substances 0.000 claims description 39
- 239000013598 vector Substances 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000001228 spectrum Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 7
- 238000009499 grossing Methods 0.000 claims description 7
- 230000001502 supplementing effect Effects 0.000 claims description 2
- 230000007774 longterm Effects 0.000 abstract description 2
- 230000000875 corresponding effect Effects 0.000 description 11
- 239000012634 fragment Substances 0.000 description 2
- 238000011065 in-situ storage Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 241000283153 Cetacea Species 0.000 description 1
- 241001125840 Coryphaenidae Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/21—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being power information
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/45—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of analysis window
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses an automatic detection and extraction method for whistle signals of Chinese white dolphin, which comprises the following steps of: preprocessing the collected data containing the whistle signal of the Chinese white dolphin; step 2: intercepting the obtained data containing the Chinese white dolphin whistle signal, cutting the data into a plurality of time slices, and executing the following operations on all the time slices one by one to obtain a signal start-stop time result containing the Chinese white dolphin whistle signal; step 3: and (3) according to the signal start-stop time result containing the Chinese white dolphin whistle signal automatically detected and stored in the step (2), cutting the signal preprocessed in the step (1), generating a time-frequency diagram for the signal containing the Chinese white dolphin whistle signal obtained after cutting, and automatically storing the time-frequency diagram generated by each section of signal. The method and the device realize full-automatic detection and extraction of mass data obtained by long-term sonar signal acquisition in dolphin research, reduce the dependence on manual intervention and improve the efficiency.
Description
Technical Field
The invention belongs to the fields of aquatic biology, hydroacoustics and hydroacoustics signal processing, and relates to an automatic detection and extraction method for whistle signal of a Chinese white dolphin.
Background
Under the condition that genetic or morphological data cannot be obtained, whistle sounds of the Chinese white dolphins can reveal relations among populations in different areas, independent populations are determined, and especially the geographically adjacent Chinese white dolphins are important for predicting population conditions and adjusting protection of the populations, so that whistle signals are valued by vast researchers. However, the method for collecting and rapidly extracting high-quality white dolphin whistle signals from the vast sea has great difficulty at present, and mainly comprises the steps that the ratio of the signals containing white dolphin whistle signals in all collected signals is very small, and the picking up of the signals is carried out on massive signals mainly by manual means, so that the cost of labor and the cost of time are both great. Therefore, if the signals collected in the sailing and in-situ mode can be analyzed by using an automatic means, fragments with high-quality white dolphin whistle signals can be automatically detected and extracted from the signals, and the research efficiency is greatly improved.
Disclosure of Invention
Aiming at the prior art, the technical problem to be solved by the invention is to provide an automatic detection and extraction method for Chinese white dolphin whistle signals, which is used for automatically detecting and extracting the Chinese white dolphin whistle signals from massive sonar data, automatically analyzing signals acquired in a sailing and in-situ mode, automatically detecting and extracting fragments with high-quality Chinese white dolphin whistle signals, improving the working efficiency and solving the problem that the massive sonar signals are screened only by manual means at present.
In order to solve the technical problems, the automatic detection and extraction method for the whistle signal of the Chinese white dolphin comprises the following steps:
Step 1: preprocessing the collected data containing the whistle signal of the Chinese white dolphin;
Step 2: intercepting the data containing the Chinese white dolphin whistle signal obtained in the step 1, cutting the data into a plurality of time slices, and executing the following operations on all the time slices one by one to obtain a signal start-stop time result containing the Chinese white dolphin whistle signal:
step A, carrying out short-time power spectrum calculation on time slices frame by frame to form a two-dimensional time-frequency diagram;
and (B) step (B): for a two-dimensional time-frequency diagram of time slices, setting a relevant sample length CorrNum and a time window length SegNum, wherein SegNum > CorrNum, corrNum is satisfied, each sampling point is calculated to be related to the subsequent point, segNum is the length of each time slice divided into a plurality of time windows, the time slices are divided into a plurality of time windows according to the time window length SegNum, and the following operation is performed on each time window: performing cross-correlation calculation on time-frequency diagrams corresponding to all sampling points in a time window to obtain a cross-correlation matrix secCoord sup, and summing the matrix secCoord sup along the row direction to obtain an integral column vector in the time window;
Step C: solving a cross-correlation integral curve of the time slices;
step D: judging whether a whistle signal of the Chinese white dolphin exists in a cross-correlation integral curve of the time segment according to a set threshold value, and obtaining the starting and ending time of the signal;
step 3: and (3) according to the signal start-stop time result containing the Chinese white dolphin whistle signal automatically detected and stored in the step (2), cutting the signal preprocessed in the step (1), generating a time-frequency diagram of the signal containing the Chinese white dolphin whistle signal obtained after cutting according to the method in the step (2), and automatically storing the time-frequency diagram generated by each section of signal.
Further, the preprocessing in step 1 includes: downsampling, high-pass filtering, and time-domain smoothing.
Further, in the step B, performing cross-correlation computation on the time-frequency diagram corresponding to all the sampling points in the time window to obtain a cross-correlation matrix secCoord sup includes:
And carrying out cross-correlation calculation on power spectrums corresponding to all sampling points in a time window:
wherein ρ XY is the cross-correlation value of the X-th column and the Y-th column, cov (X, Y) represents the covariance of the X-th column and the Y-th column of the time-frequency diagram, σ X represents the standard deviation of the X-th column of the time-frequency diagram, and σ Y represents the standard deviation of the Y-th column of the time-frequency diagram; ρ XY is used as the elements of the X line and Y column of the cross-correlation matrix secCoord to obtain a cross-correlation matrix secCoord, wherein the cross-correlation matrix secCoord is a SegNum-dimensional square matrix; taking the upper triangular matrix secCoord utri of secCoord, reserving the first CorrNum elements from the secCoord utri main diagonal elements for each row, setting 0 at all other positions, and making q= SegNum-CorrNum +1, and supplementing i elements ρ (q+i)(SegNum+a) from the q+i to obtain a matrix secCoord sup in each row, wherein i=1, 2,3, …, corrNum-1, a=1, 2, …, i.
Further, the step C of obtaining the cross-correlation integral curve of the time slices includes:
Integrating column vectors of all time windows in a time segment are sequentially connected to form a column vector matrix, an mth row element in the column vector matrix represents a cross-correlation value corresponding to an mth sampling point in the time segment, then the column vector matrix is transposed, a two-dimensional coordinate system taking time as an abscissa axis and the cross-correlation value as an ordinate axis is established, sampling time of the sampling points is taken as an abscissa, and the cross-correlation value of the sampling points is taken as an ordinate, so that a cross-correlation integral curve of each sampling point of the time segment is obtained.
Further, in the step D, judging whether the whistle signal of the white dolphin exists in the cross-correlation integral curve of the time segment according to the set threshold value, and calculating the start-stop time of the whistle signal comprises:
when the cross-correlation value of the sampling points is larger than a set threshold value, judging that the sampling points contain Chinese white dolphin whistle signals, wherein the sampling time corresponding to the sampling point which is closest to the sampling point and the cross-correlation value coincides with the threshold value in the sampling points in front of the sampling points is the starting time of the Chinese white dolphin whistle signals, and recording as t start;
when the cross-correlation value of the sampling points is smaller than a set threshold value, judging that the whistle signal of the white dolphin in the sampling points disappears, wherein the sampling time corresponding to the sampling point which is closest to the sampling point and the cross-correlation value coincides with the threshold value in the sampling points in front of the sampling points is the termination time of the whistle signal of the white dolphin, and the termination time is recorded as t end;
When the time segment contains the starting time of the Chinese white dolphin whistle signal but does not contain the ending time of the Chinese white dolphin whistle signal, recording the starting time of the Chinese white dolphin whistle signal, and recording the ending time of the Chinese white dolphin whistle signal when the next time segment.
The invention has the beneficial effects that: compared with click signals and other noise signals, the whistle signal of the Chinese white dolphin has the characteristics of obvious periodicity and strong correlation, and interference noise generated by different mechanisms is removed in a proper mode, so that the whistle signal characteristics are highlighted, the detection precision is improved, and full-automatic detection and extraction are realized. The method can realize full-automatic detection and extraction of mass data obtained by long-term sonar signal acquisition in dolphin research, reduce the dependence on manual intervention and improve the efficiency.
Drawings
FIG. 1 is an overall flow diagram of the present invention;
FIG. 2 is an automatically extracted Chinese white dolphin whistle sound result;
FIG. 3 is an automatically extracted Chinese white dolphin whistle preservation result.
Detailed Description
The invention is further described below with reference to the drawings and examples.
Because the whistle observation signal of the whale animal is not actually collected at every moment, a proper whistle signal detection method is needed, the whistle signal position is detected and determined from the observation signal, and the detected signal position is used for re-segmentation and time-frequency diagram generation. Referring to fig. 1, the present invention comprises the steps of:
Step 1: pretreatment of whistle signals of Chinese white dolphin. The collected data containing the whistle signal of the Chinese white dolphin is subjected to downsampling treatment, so that the processing speed is improved; aiming at the frequency domain characteristics of whistle signals of the Chinese white dolphins, a proper frequency band is selected to carry out high-pass filtering on the down-sampled signals, clutter which is obviously lower than the frequency band of the whistle signals is filtered, interference of ocean background noise is removed, and the signal to noise ratio is improved; aiming at the characteristic that Chinese white dolphin can send whistle and click signals (echo positioning signal strings) simultaneously in shallow sea, least square smoothing filtering is used for carrying out time domain smoothing filtering on the signals, interference of the click signals and similar pulse signals is removed, interference of short-time pulse signals on whistle frequency spectrum analysis is reduced, and signal to noise ratio is improved.
Step 2: and (5) automatically detecting whistle signals of the Chinese white dolphin. Aiming at the condition that the duration time of the Chinese white dolphin whistle signal has a certain range, intercepting data containing the Chinese white dolphin whistle signal, dividing the data into a plurality of time slices, and carrying out short-time power spectrum calculation on 1 time slice per time frame by frame (the frame refers to each time point) to form a two-dimensional time-frequency diagram; according to the characteristic that each frame of power spectrum of Chinese white dolphin whistle in one period is continuous and has strong correlation with each other, for the two-dimensional time-frequency diagram of each time segment, a correlation sample length and a time window length (the time segment length is larger than the time window length and is larger than the correlation sample length) are set, the correlation sample length is used for determining the number of subsequent points which need to be calculated and correlated with each sampling point, the time window length is used for determining the length that each time segment is divided into a plurality of time windows, and then cross-correlation calculation is carried out on the time-frequency diagram corresponding to all sampling points in the time window one by one to obtain a cross-correlation matrix. To simplify the calculation, the triangular matrix is taken, the reserved length of each row along the diagonal direction is equal to the length of the relevant sample, and all other positions are set to 0. Since the upper triangular matrix is the case where the reserved length is smaller than the relevant sample length, the corresponding missing value is complemented. And summing the matrix after the supplementary values along the row direction to obtain an integral column vector in a time window. Repeating the steps for each time window, integrating the cross-correlations of all the time windows to form a new matrix, wherein the vector is the cross-correlation value corresponding to all the sampling points in the current time segment, namely the first row represents the cross-correlation value of the first sampling point, then transposing the vector, and sequentially drawing the cross-correlation value on a two-dimensional coordinate system according to the sequence of each sampling point to obtain the cross-correlation integral curve of each time point of the current time segment.
Correlation integral threshold analysis. Judging a cross-correlation integral curve formed by the correlation integral of the current time slice according to time points by setting a reasonable threshold value, automatically judging whether whistle signals of dolphins exist in the cross-correlation integral curve, calculating the starting and ending time of the whistle signals, judging that whistle signals exist in the current time point if the cross-correlation integral result of a certain time point is larger than the threshold value, and judging that the whistle signals exist in the current time point and the coincident point before the current time point is larger than the threshold value as the starting time of the whistle signals; and if the cross-correlation integral result at a certain time point is smaller than the threshold value, judging that the whistle signal of the dolphin at the current time point disappears, and if the cross-correlation integral result at the certain time point is smaller than the threshold value, judging that the coincident point before the time point is smaller than the threshold value is the termination time of the whistle signal. Since each slicing time is 1-2 seconds, the whistle of the white dolphin in the current time segment can be always in a condition of no time, at the moment, only the starting point of the whistle is recorded, the whistle end point is recorded in the next time segment according to the duration range of the whistle signal of the white dolphin being 60-1972ms, and the starting and ending time is recorded and stored finally.
And then carrying out the steps on each time segment to obtain the start-stop time of all signals.
Step 3: and (5) automatically extracting whistle signals of the Chinese white dolphin.
And (3) according to the starting and stopping time results which are automatically detected and stored in the step (2) and contain the whistle signals of the Chinese white dolphin, segmenting all the time signals again, generating a time-frequency diagram for the signals by using the parameters in the step (2), and automatically storing the generated time-frequency diagram of each section of signal.
Examples are given below in connection with specific parameters.
Referring to fig. 1, the present invention includes the steps of:
Step 1: pretreatment of whistle signals of Chinese white dolphin.
The waveform signal is downsampled, so that the processing speed is improved. As the whistle signal frequency range of the white dolphin (Hainan area of China) is 0.17-15.80kHz, the signal sampling rate is firstly reduced to 36-42kHz, and the data dataRaw _ds are obtained.
The waveform signal is noise reduced, and the signal-to-noise ratio and the detection precision are improved. In the signal acquisition process, a large amount of background noise exists in a frequency band below 700Hz, and the fundamental frequency range of the whistle signal of the Chinese dolphin is 0.71-17.30kHz, so that the time sequence signal dataRaw _ds is subjected to high-pass filtering, the passband frequency fpass is set to 700Hz, and the filtered signal data_ hipass is obtained. Aiming at click signals mixed in whistle signals of Chinese white dolphins, the frequency bands and the intensity of the click signals can generate obvious interference on signal detection, but the click signals belong to typical pulse signals, time domain smoothing can be used for noise reduction, in the patent, data_ hipass is subjected to time domain-based least square smoothing filtering, and a first-order Savitzky-Golay filter is used for smoothing, so that a smoothed time sequence signal data_dn is obtained.
Step 2: and (5) automatically detecting whistle signals of the Chinese white dolphin.
Aiming at the duration range of the whistle signal of the Chinese white dolphin is 60-1972ms, intercepting the observed signal after noise reduction, dividing the observed signal into a plurality of time segments with the length of 2s, setting the signal length as Ns, and setting the sampling rate as Fs, namely each time segment comprises 2Fs sampling points; then 1 time segment at a time: s i (n), n=1, 2,3,..2 Fs,The fourier transform results are expressed as X i (k), k=1, 2,3,..2 Fs,/>Recalculating the power spectrum of the signalk＝1,2,3,...,2Fs,/>A power spectrum of 0-20kHz is calculated and a time-frequency diagram is generated by applying a smooth Hanning window, 680-point FFT, 98% window overlap ratio, and 500Hz frequency resolution to the time segment.
A cross-correlation matrix is calculated. For the time-frequency diagram in the time segment, a correlation sample length CorrNum and a time window length SegNum (SegNum > CorrNum) are set, corrNum are used for determining the number of subsequent points related to each sampling point to be calculated, segNum are used for determining the length of the time window. Sequentially taking SegNum sampling point sets secreta (j), j=1,..m, m=2fs/SegNum; and then carrying out cross-correlation calculation on power spectrums corresponding to all sampling points in SecMeta (j):
Wherein ρ XY represents the Pearson correlation coefficient, which is the cross-correlation value of the X-th column and the Y-th column of the time-frequency chart, cov (X, Y) represents the covariance of the X-th column and the Y-th column of the time-frequency chart, and σ X represents the standard deviation of the X-th column of the time-frequency chart. Then, the cross-correlation matrix secCoord of the SecMeta set is obtained, wherein secCoord is a SegNum-dimensional square matrix, and secCoord matrix is:
The cross-correlation matrix simplifies the processing. secCoord is a symmetric matrix, and for simplicity of calculation, the upper triangular matrix is taken and denoted secCoord utri:
Since each calculation only concerns the correlation of CorrNum samples adjacent to each sample, matrix secCoord utri retains only the first CorrNum values for each row in the diagonal direction, with all points located thereafter set to 0, denoted secCoord trap:
However, when CorrNum +q-1= SegNum, i.e., q= SegNum-CorrNum +1, each row after the q-th row (without q rows) is less than CorrNum values (total CorrNum-1 rows are less than CorrNum values), so the q+1th row is complemented by a value ρ (q+1)(SegNum+1), the q+2th row is complemented by two values ρ (q+2)(SegNum+1) and ρ (q+2)(SegNum+2), and so on, the SegNum th row (last row) is complemented by CorrNum-1 values. The complemented matrix is noted secCoord sup:
A cross-correlation integral curve is generated. For matrix secCoord sup, it is summed in the row direction to obtain an integral vector The above steps are then repeated for each time window, i.e. for each secmata (j), j=1, 2,3,..2 Fs/SegNum the integration vectors CCAvg (j) are calculated and all CCAvg (j) are connected in order, resulting in the total correlation integration vector allcavg (i) for the ith time segment, the dimension of allcavg (i) being 1 column, 2Fs row. The vector is the cross-correlation value corresponding to all sampling points in the current time segment, namely, the first row represents the cross-correlation value of the first sampling point, then the vector is transposed, and the cross-correlation values are sequentially drawn on a two-dimensional coordinate system according to the sequence of each sampling point, so that the cross-correlation integral curve of each time point of the current time segment is obtained.
Correlation integral threshold analysis. Judging a cross-correlation integral curve formed by total correlation integral vector AllCCAvg (i) according to time points by setting a reasonable threshold shd, automatically judging whether a whistle signal of a dolphin exists in the cross-correlation integral curve according to the set threshold, calculating the starting and ending time of the whistle signal, judging that the whistle signal exists at the current time point if the cross-correlation integral result of one time point is larger than the threshold, and marking the coincident point before the time point is larger than the threshold as the starting time of the whistle signal as t start; if the cross-correlation integral result of one time point is smaller than the threshold value, judging that the whistle signal of the dolphin at the current time point disappears, and if the cross-correlation integral result of the time point is smaller than the threshold value, the coincident point before the threshold value is the termination time of the whistle signal, and is marked as t end, and since the slicing time is 2 seconds each time, the whistle of the dolphin in the current time segment can appear without the end, at the moment, only the whistle starting point is recorded, and the whistle ending point is recorded when the next time segment is processed according to the duration range of the whistle signal of the dolphin in the Chinese time segment being 60-1972 ms. Then, the steps are carried out on each time segment to obtain the start and stop time of all time signals, and the start and stop time is recorded and stored as
Step 3: and (5) automatically extracting whistle signals of the Chinese white dolphin.
And (3) according to the starting and stopping time results which are automatically detected and stored in the step (2) and contain the whistle signal of the Chinese dolphin, cutting all time signals again, generating a time-frequency diagram for the n sections of signals by using the parameters in the step (2), and automatically storing the generated time-frequency diagram of each section of signals, wherein the stored signals are shown in figures 2 and 3.
Claims (5)
1. The automatic detection and extraction method for the whistle signal of the Chinese white dolphin is characterized by comprising the following steps of:
Step 1: preprocessing the collected data containing the whistle signal of the Chinese white dolphin;
Step 2: intercepting the data containing the Chinese white dolphin whistle signal obtained in the step 1, cutting the data into a plurality of time slices, and executing the following operations on all the time slices one by one to obtain a signal start-stop time result containing the Chinese white dolphin whistle signal:
step A, carrying out short-time power spectrum calculation on time slices frame by frame to form a two-dimensional time-frequency diagram;
and (B) step (B): for a two-dimensional time-frequency diagram of time slices, setting a relevant sample length CorrNum and a time window length SegNum, wherein SegNum > CorrNum, corrNum is satisfied, each sampling point is calculated to be related to the subsequent point, segNum is the length of each time slice divided into a plurality of time windows, the time slices are divided into a plurality of time windows according to the time window length SegNum, and the following operation is performed on each time window: performing cross-correlation calculation on time-frequency diagrams corresponding to all sampling points in a time window to obtain a cross-correlation matrix secCoord sup, and summing the matrix secCoord sup along the row direction to obtain an integral column vector in the time window;
Step C: solving a cross-correlation integral curve of the time slices;
step D: judging whether a whistle signal of the Chinese white dolphin exists in a cross-correlation integral curve of the time segment according to a set threshold value, and obtaining the starting and ending time of the signal;
step 3: and (3) according to the signal start-stop time result containing the Chinese white dolphin whistle signal automatically detected and stored in the step (2), cutting the signal preprocessed in the step (1), generating a time-frequency diagram of the signal containing the Chinese white dolphin whistle signal obtained after cutting according to the method in the step (2), and automatically storing the time-frequency diagram generated by each section of signal.
2. The automatic detection and extraction method for whistle signals of Chinese white dolphin according to claim 1, which is characterized in that: the pretreatment in the step 1 comprises the following steps: downsampling, high-pass filtering, and time-domain smoothing.
3. The automatic detection and extraction method for whistle signals of Chinese white dolphin according to claim 1, which is characterized in that: and step B, performing cross-correlation computation on the time-frequency graphs corresponding to all the sampling points in the time window to obtain a cross-correlation matrix secCoord sup includes:
And carrying out cross-correlation calculation on power spectrums corresponding to all sampling points in a time window:
wherein ρ XY is the cross-correlation value of the X-th column and the Y-th column, cov (X, Y) represents the covariance of the X-th column and the Y-th column of the time-frequency diagram, σ X represents the standard deviation of the X-th column of the time-frequency diagram, and σ Y represents the standard deviation of the Y-th column of the time-frequency diagram; ρ XY is used as the elements of the X line and Y column of the cross-correlation matrix secCoord to obtain a cross-correlation matrix secCoord, wherein the cross-correlation matrix secCoord is a SegNum-dimensional square matrix; taking the upper triangular matrix secCoord utri of secCoord, reserving the first CorrNum elements from the secCoord utri main diagonal elements for each row, setting 0 at all other positions, and making q= SegNum-CorrNum +1, and supplementing i elements ρ (q+i)(SegNum+a) from the q+i to obtain a matrix secCoord sup in each row, wherein i=1, 2,3, …, corrNum-1, a=1, 2, …, i.
4. The automatic detection and extraction method for whistle signals of Chinese white dolphin according to claim 1, which is characterized in that: the step C of calculating the cross-correlation integral curve of the time slices comprises the following steps:
Integrating column vectors of all time windows in a time segment are sequentially connected to form a column vector matrix, an mth row element in the column vector matrix represents a cross-correlation value corresponding to an mth sampling point in the time segment, then the column vector matrix is transposed, a two-dimensional coordinate system taking time as an abscissa axis and the cross-correlation value as an ordinate axis is established, sampling time of the sampling points is taken as an abscissa, and the cross-correlation value of the sampling points is taken as an ordinate, so that a cross-correlation integral curve of each sampling point of the time segment is obtained.
5. The automatic detection and extraction method for whistle signals of Chinese white dolphin according to claim 1, which is characterized in that: and D, judging whether a Chinese white dolphin whistle signal exists in a cross-correlation integral curve of the time segment according to the set threshold value, and calculating the starting and ending time of the Chinese white dolphin whistle signal comprises the following steps:
when the cross-correlation value of the sampling points is larger than a set threshold value, judging that the sampling points contain Chinese white dolphin whistle signals, wherein the sampling time corresponding to the sampling point which is closest to the sampling point and the cross-correlation value coincides with the threshold value in the sampling points in front of the sampling points is the starting time of the Chinese white dolphin whistle signals, and recording as t start;
when the cross-correlation value of the sampling points is smaller than a set threshold value, judging that the whistle signal of the white dolphin in the sampling points disappears, wherein the sampling time corresponding to the sampling point which is closest to the sampling point and the cross-correlation value coincides with the threshold value in the sampling points in front of the sampling points is the termination time of the whistle signal of the white dolphin, and the termination time is recorded as t end;
When the time segment contains the starting time of the Chinese white dolphin whistle signal but does not contain the ending time of the Chinese white dolphin whistle signal, recording the starting time of the Chinese white dolphin whistle signal, and recording the ending time of the Chinese white dolphin whistle signal when the next time segment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210541058.1A CN115050386B (en) | 2022-05-17 | 2022-05-17 | Automatic detection and extraction method for whistle signal of Chinese white dolphin |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210541058.1A CN115050386B (en) | 2022-05-17 | 2022-05-17 | Automatic detection and extraction method for whistle signal of Chinese white dolphin |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115050386A CN115050386A (en) | 2022-09-13 |
| CN115050386B true CN115050386B (en) | 2024-05-28 |
Family
ID=83159591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210541058.1A Active CN115050386B (en) | 2022-05-17 | 2022-05-17 | Automatic detection and extraction method for whistle signal of Chinese white dolphin |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115050386B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116660876A (en) * | 2023-05-26 | 2023-08-29 | 吉林大学 | Automatic detection and positioning method for echo positioning signals of Chinese white dolphin |
| CN117031402B (en) * | 2023-10-10 | 2023-12-12 | 南方海洋科学与工程广东省实验室(珠海) | Acousto-optic fusion monitoring equipment and method based on small buoy |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0279099A (en) * | 1988-09-14 | 1990-03-19 | Nec Corp | Multi-pulse voice processor |
| CN101650430A (en) * | 2008-12-26 | 2010-02-17 | 中国科学院声学研究所 | System for acquiring dolphin's sound signal and method thereof |
| KR20110010407A (en) * | 2009-07-24 | 2011-02-01 | 세종대학교산학협력단 | Signal checking and pattern information acquisition apparatus and method thereof |
| CN102606891A (en) * | 2012-04-11 | 2012-07-25 | 广州东芝白云自动化***有限公司 | Water leakage detector, water leakage detecting system and water leakage detecting method |
| CN102682765A (en) * | 2012-04-27 | 2012-09-19 | 中咨泰克交通工程集团有限公司 | Expressway audio vehicle detection device and method thereof |
| CN107024718A (en) * | 2017-05-31 | 2017-08-08 | 西南石油大学 | Poststack earthquake fluid Forecasting Methodology based on CEEMD SPWVD Time-frequency Spectrum Analysis |
| CN107533839A (en) * | 2015-12-17 | 2018-01-02 | 华为技术有限公司 | A kind of processing method and equipment to surrounding environment sound |
| CN107635181A (en) * | 2017-09-15 | 2018-01-26 | 哈尔滨工程大学 | A kind of multiple access based on channel study perceives the feedback optimized method in source |
| CN110211596A (en) * | 2019-05-29 | 2019-09-06 | 哈尔滨工程大学 | One kind composing entropy cetacean whistle signal detection method based on Mel subband |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5380362B2 (en) * | 2010-05-17 | 2014-01-08 | パナソニック株式会社 | Quality inspection method and quality inspection apparatus |
| JP2014215461A (en) * | 2013-04-25 | 2014-11-17 | ソニー株式会社 | Speech processing device, method, and program |
-
2022
- 2022-05-17 CN CN202210541058.1A patent/CN115050386B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0279099A (en) * | 1988-09-14 | 1990-03-19 | Nec Corp | Multi-pulse voice processor |
| CN101650430A (en) * | 2008-12-26 | 2010-02-17 | 中国科学院声学研究所 | System for acquiring dolphin's sound signal and method thereof |
| KR20110010407A (en) * | 2009-07-24 | 2011-02-01 | 세종대학교산학협력단 | Signal checking and pattern information acquisition apparatus and method thereof |
| CN102606891A (en) * | 2012-04-11 | 2012-07-25 | 广州东芝白云自动化***有限公司 | Water leakage detector, water leakage detecting system and water leakage detecting method |
| CN102682765A (en) * | 2012-04-27 | 2012-09-19 | 中咨泰克交通工程集团有限公司 | Expressway audio vehicle detection device and method thereof |
| CN107533839A (en) * | 2015-12-17 | 2018-01-02 | 华为技术有限公司 | A kind of processing method and equipment to surrounding environment sound |
| CN107024718A (en) * | 2017-05-31 | 2017-08-08 | 西南石油大学 | Poststack earthquake fluid Forecasting Methodology based on CEEMD SPWVD Time-frequency Spectrum Analysis |
| CN107635181A (en) * | 2017-09-15 | 2018-01-26 | 哈尔滨工程大学 | A kind of multiple access based on channel study perceives the feedback optimized method in source |
| CN110211596A (en) * | 2019-05-29 | 2019-09-06 | 哈尔滨工程大学 | One kind composing entropy cetacean whistle signal detection method based on Mel subband |
Non-Patent Citations (4)
| Title |
|---|
| 厦门海域中华白海豚定位click信号特性;牛富强;杨燕明;许肖梅;文洪涛;刘贞文;;厦门大学学报(自然科学版);20120928(第05期);全文 * |
| 基于单类支持向量机的异常声音检测;陈志全;杨骏;乔树山;;电子设计工程;20161205(第23期);全文 * |
| 基于海豚whistle信号的仿生主动声纳隐蔽探测技术研究;殷敬伟;刘强;陈阳;朱广平;生雪莉;;兵工学报;20160515(第05期);全文 * |
| 宽吻海豚Click信号的时频滤波检测方法;杨武夷;孙馨喆;宋忠长;张宇;杨燕明;;声学学报;20170715(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115050386A (en) | 2022-09-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115050386B (en) | Automatic detection and extraction method for whistle signal of Chinese white dolphin | |
| Mellinger et al. | Recognizing transient low-frequency whale sounds by spectrogram correlation | |
| CN108814590B (en) | Detection method of electrocardio QRS wave group and electrocardio analysis method thereof | |
| CN115206334B (en) | Polar region acoustic signal enhancement and identification method | |
| CN114705426A (en) | Early fault diagnosis method for rolling bearing | |
| CN111414832A (en) | Real-time online recognition and classification system based on whale dolphin low-frequency underwater acoustic signals | |
| CN114782745B (en) | Ocean sound velocity profile classification method and device based on machine learning | |
| Lin et al. | Automatic classification of delphinids based on the representative frequencies of whistles | |
| CN115061203A (en) | Mine single-channel microseismic signal noise reduction method based on frequency domain singular value decomposition and application | |
| Roch et al. | Using context to train time-domain echolocation click detectors | |
| CN110236538B (en) | Automatic real-time detection method for motion starting point based on electromyographic signal | |
| McCullough et al. | Acoustic classification of false killer whales in the Hawaiian islands based on comprehensive vocal repertoire | |
| CN1216361C (en) | Estimating the pitch of a speech signal using a binary signal | |
| CN109377982A (en) | A kind of efficient voice acquisition methods | |
| CN102783946B (en) | Automatic brain source locating method and device | |
| CN110335613B (en) | Bird identification method adopting pickup for real-time detection | |
| Crudeli et al. | Abnormal carbonate diagenesis in Holocene–late Pleistocene sapropel-associated sediments from the Eastern Mediterranean; evidence from Emiliania huxleyi coccolith morphology | |
| CN114114400B (en) | Microseism event effective signal pickup method | |
| CN115169644A (en) | Energy consumption prediction method and system for dredger | |
| CN114882901A (en) | Shrimp acoustic signal time-frequency feature extraction method based on frequency domain convolution and marginal spectrum feedback | |
| CN114638265A (en) | Milling flutter discrimination method based on signal convolution neural network | |
| CN109065070B (en) | Kernel function-based audio characteristic signal dimension reduction method | |
| CN109602414B (en) | Multi-view-angle conversion electrocardiosignal data enhancement method | |
| CN112086105A (en) | Target identification method based on Gamma atom sub-band continuous spectrum characteristics | |
| CN111352156B (en) | Multiple independent seismic data automatic velocity analysis method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |